Automated training data generation for microscopy focus classification.

Area of Science:

• Digital pathology
• Microscopy imaging
• Computational pathology

Background:

• Accurate focus quality is critical for pathologists to reliably interpret digital microscopy scans.
• Current methods for training focus assessment classifiers are hindered by the cost and time associated with manual data annotation.

Purpose of the Study:

• To develop an automated method for generating large-scale training datasets for focus quality classifiers.
• To evaluate the performance of a focus quality classifier trained on automatically generated data compared to manually annotated data.

Main Methods:

• Utilized image stacks to automatically generate extensive training data for a focus quality classifier.
• Extracted a comprehensive set of image features to train the classifier.
• Experimentally validated the classifier's performance using both automatically generated and manually annotated datasets.

Main Results:

• The classifier trained with automatically generated data achieved performance comparable to that trained with manually annotated data.
• The developed classifier accurately identifies out-of-focus regions in microscopy scans.
• The system provides valuable focus quality feedback to users and can detect microscopy design issues.

Conclusions:

• Automated training data generation using image stacks is a viable and efficient alternative to manual annotation for digital microscopy focus assessment.
• The developed classifier enhances the reliability of digital pathology by ensuring accurate focus quality evaluation.
• This approach has the potential to improve microscopy system design and user feedback mechanisms.